diff --git a/_primitives/README.md b/_primitives/README.md
new file mode 100644
index 0000000..7383e19
--- /dev/null
+++ b/_primitives/README.md
@@ -0,0 +1,30 @@
+# \_primitives — first-class building blocks
+
+`_primitives/` holds standalone utilities that agents, hooks, and skills
+(including `/compose-solution`) depend on. Unlike `_blocks/` (behavioral
+markdown) or `_manifests/` (agent TOML), primitives are executable shell
+programs installed at `$HOME/.claude/agents/_primitives/` by `install.sh`.
+
+## Current primitives
+
+| Primitive | Purpose | Invocation |
+|---|---|---|
+| `tomd.sh` | Universal non-native-format → markdown converter (PDF, DOCX, XLSX, PPTX, CSV, images, code). | `~/.claude/agents/_primitives/tomd.sh <file>` |
+
+`tomd.sh` is ported from [KeiAgent](https://…/KeiAgent) `bin/keiagent-tomd` —
+same format matrix, KeiSeiKit-style error tags (`[tomd]`), configurable
+cache directory (`KEISEI_TOMD_CACHE`).
+
+## Hook integration
+
+`hooks/tomd-preread.sh` is a PreToolUse(Read) hook that auto-redirects
+Claude to the converted markdown when a Read targets `.docx / .doc / .xlsx /
+.pptx / .csv`. Cached under `$KEISEI_TOMD_CACHE` (default
+`/tmp/keisei-tomd-cache`).
+
+## `/compose-solution` discovery
+
+Phase 3 prior-art sweep greps `_primitives/` alongside `_blocks/`,
+`_manifests/`, `skills/`, `_bridges/`, `hooks/`. If a user task involves
+file-format parsing, the meta-composer surfaces `tomd` automatically —
+reuse over rewrite (RULE "No Patching").
diff --git a/_primitives/tomd.sh b/_primitives/tomd.sh
new file mode 100755
index 0000000..7182f7d
--- /dev/null
+++ b/_primitives/tomd.sh
@@ -0,0 +1,180 @@
+#!/usr/bin/env bash
+# tomd — universal non-native-format → markdown converter.
+# Ported from ~/Projects/KeiAgent/bin/keiagent-tomd. First-class primitive.
+# Install path: $HOME/.claude/agents/_primitives/tomd.sh.
+# Deps: pandoc, python3, jq. Optional: pymupdf4llm, openpyxl, tesseract.
+
+set -euo pipefail
+
+check_deps() {
+  local missing=()
+  command -v pandoc  >/dev/null 2>&1 || missing+=("pandoc  (brew install pandoc)")
+  command -v python3 >/dev/null 2>&1 || missing+=("python3 (system)")
+  command -v jq      >/dev/null 2>&1 || missing+=("jq      (brew install jq)")
+  if [ "${#missing[@]}" -gt 0 ]; then
+    echo "[tomd] missing prerequisites:" >&2
+    for m in "${missing[@]}"; do echo "[tomd]   - $m" >&2; done
+    echo "[tomd] hint: brew install pandoc jq && pip3 install pymupdf4llm openpyxl" >&2
+    exit 1
+  fi
+}
+
+detect_format() {
+  local f="$1"
+  [ "$f" = "-" ] && { echo stdin; return; }
+  local l; l=$(printf '%s' "$f" | tr '[:upper:]' '[:lower:]')
+  case "$l" in
+    *.pdf)  echo pdf ;; *.docx) echo docx ;; *.doc) echo doc ;;
+    *.html|*.htm) echo html ;; *.pptx) echo pptx ;;
+    *.xlsx) echo xlsx ;; *.csv)  echo csv ;;
+    *.json) echo "fence:json" ;; *.yaml|*.yml) echo "fence:yaml" ;;
+    *.xml)  echo "fence:xml" ;; *.toml) echo "fence:toml" ;; *.sql) echo "fence:sql" ;;
+    *.png|*.jpg|*.jpeg|*.gif|*.webp|*.svg) echo image ;;
+    *.py)   echo "fence:python" ;; *.go) echo "fence:go" ;;
+    *.ts|*.tsx) echo "fence:typescript" ;; *.js|*.jsx) echo "fence:javascript" ;;
+    *.rs)   echo "fence:rust" ;; *.c|*.h) echo "fence:c" ;; *.cpp|*.hpp) echo "fence:cpp" ;;
+    *.swift) echo "fence:swift" ;; *.sh|*.bash|*.zsh) echo "fence:bash" ;;
+    *.zig)  echo "fence:zig" ;; *.md) echo md ;; *) echo text ;;
+  esac
+}
+
+convert_pdf() {
+  python3 - "$1" <<'PYEOF'
+import sys
+p=sys.argv[1]
+try: import pymupdf4llm; print(pymupdf4llm.to_markdown(p))
+except ImportError:
+    try:
+        import fitz; doc=fitz.open(p)
+        for page in doc: print(page.get_text("text")); print()
+    except ImportError:
+        sys.stderr.write("[tomd] pdf: pip3 install pymupdf4llm\n"); sys.exit(1)
+PYEOF
+}
+
+convert_pandoc() {
+  local from="${2:-}"
+  if [ -n "$from" ]; then pandoc -f "$from" -t markdown --wrap=none "$1"
+  else pandoc -t markdown --wrap=none "$1"; fi
+}
+
+convert_doc() {
+  if ! command -v textutil >/dev/null 2>&1; then
+    echo "[tomd] .doc: textutil not available (macOS only). Convert to .docx first." >&2
+    exit 1
+  fi
+  local tmp; tmp=$(mktemp /tmp/tomd-XXXX.html)
+  textutil -convert html -output "$tmp" "$1"
+  pandoc -f html -t markdown --wrap=none "$tmp"; rm -f "$tmp"
+}
+
+convert_csv() {
+  python3 - "$1" <<'PYEOF'
+import csv, sys
+with open(sys.argv[1]) as f: rows=list(csv.reader(f))
+if not rows: sys.exit(0)
+print('| '+' | '.join(rows[0])+' |')
+print('|'+'|'.join(['---']*len(rows[0]))+'|')
+for r in rows[1:]: print('| '+' | '.join(r)+' |')
+PYEOF
+}
+
+convert_xlsx() {
+  python3 - "$1" <<'PYEOF'
+import sys
+try: import openpyxl
+except ImportError: sys.stderr.write("[tomd] xlsx: pip3 install openpyxl\n"); sys.exit(1)
+wb=openpyxl.load_workbook(sys.argv[1], data_only=True)
+for name in wb.sheetnames:
+    ws=wb[name]; print(f"## Sheet: {name}\n")
+    rows=list(ws.iter_rows(values_only=True))
+    if not rows: continue
+    h=[str(c) if c is not None else '' for c in rows[0]]
+    print('| '+' | '.join(h)+' |'); print('|'+'|'.join(['---']*len(h))+'|')
+    for r in rows[1:]:
+        print('| '+' | '.join(str(c) if c is not None else '' for c in r)+' |')
+    print()
+PYEOF
+}
+
+fence() { echo "\`\`\`$1"; cat "$2"; echo '```'; }
+
+convert_json() {
+  echo '```json'
+  if [ "$1" = "-" ]; then jq '.' 2>/dev/null || cat
+  else jq '.' "$1" 2>/dev/null || cat "$1"; fi
+  echo '```'
+}
+
+convert_image() {
+  echo "![$(basename "$1")]($1)"
+  if command -v tesseract >/dev/null 2>&1; then
+    echo; echo "**OCR:**"; echo '```'
+    tesseract "$1" stdout 2>/dev/null || echo "(failed)"
+    echo '```'
+  fi
+}
+
+convert_stdin() {
+  local c; c=$(cat)
+  if printf '%s' "$c" | jq '.' >/dev/null 2>&1; then
+    echo '```json'; printf '%s' "$c" | jq '.'; echo '```'
+  else printf '%s\n' "$c"; fi
+}
+
+convert_single() {
+  local f="$1" fmt; fmt=$(detect_format "$f")
+  case "$fmt" in
+    pdf)        convert_pdf "$f" ;;
+    docx)       convert_pandoc "$f" ;;
+    doc)        convert_doc "$f" ;;
+    html)       convert_pandoc "$f" html ;;
+    pptx)       convert_pandoc "$f" ;;
+    xlsx)       convert_xlsx "$f" ;;
+    csv)        convert_csv "$f" ;;
+    fence:json) convert_json "$f" ;;
+    fence:*)    fence "${fmt#fence:}" "$f" ;;
+    image)      convert_image "$f" ;;
+    md|text)    cat "$f" ;;
+    stdin)      convert_stdin ;;
+    *)          echo "[tomd] unknown format for $f" >&2; return 1 ;;
+  esac
+}
+
+convert_dir() {
+  local dir="${1:-.}" outdir="${1:-.}/_md" count=0
+  mkdir -p "$outdir"
+  while IFS= read -r -d '' file; do
+    local fmt; fmt=$(detect_format "$file")
+    { [ "$fmt" = "text" ] || [ "$fmt" = "md" ]; } && continue
+    local out="$outdir/$(basename "${file%.*}").md"
+    echo "[tomd]   $file -> $out" >&2
+    convert_single "$file" > "$out" 2>/dev/null && count=$((count+1)) || true
+  done < <(find "$dir" -maxdepth 2 -type f -not -path '*/_md/*' -not -path '*/.git/*' -print0)
+  echo "[tomd] converted $count files -> $outdir" >&2
+}
+
+usage() { cat <<'EOF'
+Usage: tomd <file> [--output out.md]
+       tomd --dir <folder>
+       cat file | tomd -
+
+Converts non-native formats to markdown (for LLM ingestion).
+Formats: PDF DOCX DOC HTML PPTX XLSX CSV JSON YAML XML TOML SQL
+         images(+OCR) code(py/go/ts/rs/c/swift/sh/zig) text
+EOF
+}
+
+check_deps
+case "${1:-}" in
+  -h|--help|help) usage; exit 0 ;;
+  --dir)          convert_dir "${2:-.}"; exit 0 ;;
+  "")             usage; exit 1 ;;
+esac
+
+output=""; file="$1"; shift
+while [ $# -gt 0 ]; do
+  case "$1" in --output) output="$2"; shift 2 ;; *) shift ;; esac
+done
+if [ -n "$output" ]; then convert_single "$file" > "$output"; echo "[tomd] wrote: $output" >&2
+else convert_single "$file"; fi